PlateLedger

Project Description

PlateLedger is a GrayWorks product for independent restaurants running one to ten locations. It reads the point of sale, multiplies every completed sale through the dish's recipe, and writes the ingredients off the shelf: three chicken quesadillas sold records six tortillas, 378 grams of chicken, and 180 grams of cheddar gone. Square and Clover connect directly and read-only — nothing is ever written back to the merchant's POS — while Toast and Shift4/SkyTab come in through a sales-mix CSV import.

From that ledger the app forecasts days of stock against each supplier's lead time and raises alerts that show their reasoning — the sales average used, what is on hand, and the lead time — rather than a bare number. Ingredients below par become a suggested order grouped by preferred supplier and rounded up to whole packs, which becomes a purchase-order draft to approve and send as HTML and PDF, then receive back into stock. A physical count taken on a phone in the walk-in reconciles the theoretical figure against the shelf and reports the variance in dollars, and six reports export to CSV.

The application is Node 22 and TypeScript on Express 5 over an append-only SQLite stock ledger, with Argon2 password hashing, session cookies or scoped API keys, org-scoped data access, and Stripe Checkout driving a 14-day trial and four paid plans. An AI layer adds demand forecasting — Google's TimesFM 2.5 running in a private Python worker behind a typed HTTP boundary — and an in-app assistant that answers questions about the signed-in org's own data and never writes without a confirmation. It ships as a non-root, multi-stage Docker image behind a TLS reverse proxy with a strict Content-Security-Policy and HSTS, scores 100 on Lighthouse accessibility on its main screens, and is covered by more than 700 vitest tests across 83 files plus Playwright end-to-end specs.